Methods, systems, and carriers for determining the characteristics of fatigue striations and the presence of fatigue damage.

Abstract

This invention relates to methods, systems, and carriers for determining the characteristics of fatigue fringes and the presence of fatigue damage. To improve the determination or assessment of fatigue damage in parts, a computer-implemented method for determining the fringe characteristics of fatigue fringes on a sample surface of a part is proposed. The sample surface is imaged using a scanning electron microscope to obtain sample images that may contain the fatigue fringes. Sample image blocks that may contain fatigue fringes are selected from the sample images for further processing. After normalizing the sample image blocks and enhancing the linear regularity contained within them, the resulting normalized image blocks are subjected to autocorrelation, Lardon transform, and spectral analysis. If any fatigue fringes are present, the resulting power spectrum of the transformed image block contains information about the fringe characteristics of the fatigue fringes contained in the sample image. Furthermore, a system for performing the method is proposed.

Description

Technical Field

[0001] This invention relates to methods, systems, and computer-readable data carriers for determining the characteristics of fatigue striations. The invention also relates to methods, systems, and computer-readable data carriers for determining the presence of fatigue damage. Background Technology

[0002] Refer to the following existing technical documents:

[0003] [1] RC Gonzales, REWoods, “Digital Image Processing”, Prentice Hall International, 2007

[0004] [2] R. Glyons, “Understanding Digital Signal Processing”, Prentice Hall, 2011

[0005] [3] Zuiderveld, Karel, “Contrast-Limited Adaptive Histogram Equalization”, Graphic Gems IV. San Diego: Academic Press Professional, 1994, 474-485.

[0006] [4] Otsu, N., “A Threshold Selection Method from Gray-Level Histograms”, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, No. 1, 1979, pp. 62-66

[0007] [5] Bailey, Donald, “Detecting regular patterns using frequency domain self-filtering”, Vol. 1, 440-443, 1997, 10.1109 / ICIP.1997.647801 Summary of the Invention

[0008] The purpose of this invention is to improve the determination or assessment of fatigue damage to parts, such as aircraft parts.

[0009] This objective is achieved by means of the features described in the independent claim. Preferred embodiments are the subject of the dependent claims.

[0010] This invention provides a computer-implemented method for determining fatigue striation features on the sample surface of a part, preferably an aircraft part, the method comprising:

[0011] Step 1.1 Use a scanning microscope to image the surface of the sample to obtain a sample image that may contain the fatigue striations;

[0012] Step 1.2 Using an image block selection device, select sample image blocks from the sample images that may contain the fatigue fringes for further processing;

[0013] Step 1.3 Using a preprocessing device, the sample image blocks are normalized and the linear regular structure contained in the sample image blocks is enhanced to obtain normalized image blocks;

[0014] Step 1.4 Use an autocorrelation device to determine the autocorrelation of the normalized image patch to obtain an autocorrelation image patch;

[0015] Step 1.5 Use the Radon transform device to perform Radon transform on the autocorrelation image block to obtain the transformed image block;

[0016] Step 1.6 Using a frequency analysis device, perform spectral analysis on the transformed image block and determine the power spectrum of the transformed image block;

[0017] Step 1.7 Using the fringe feature determination device, determine at least one fringe feature suitable for indicating the presence of fatigue fringes in the sample image block based on the power spectrum.

[0018] Preferably, in step 1.1, the scanning microscope device is selected from the group consisting of a scanning electron microscope or a scanning probe microscope.

[0019] Preferably, in step 1.2, multiple sample image blocks of different sizes are selected, and the sample image blocks are centered on the same point of the sample image, and steps 1.3 to 1.7 are performed on each of the sample image blocks.

[0020] Preferably, in step 1.3, a Gaussian filtering operation is performed on the sample image block to obtain the normalized image block.

[0021] Preferably, in step 1.3, histogram equalization is performed on the sample image block to obtain the normalized image block.

[0022] Preferably, in step 1.3, image binarization and subsequent morphological operations (such as erosion, dilation, or skeletonization) are performed on the sample image block to obtain the normalized image block.

[0023] Preferably, in step 1.3, a frequency domain self-filtering operation is performed on the sample image block to obtain the normalized image block.

[0024] Preferably, in step 1.3, a windowing operation (such as Hamming window) is performed on the sample image patch to obtain the normalized image patch.

[0025] Preferably, in step 1.3, a normalization operation is performed on the sample image block to normalize the intensity values ​​to a fixed range to obtain the normalized image block.

[0026] Preferably, in step 1.4, the autocorrelation is determined via the frequency domain using zero-padding input image blocks.

[0027] Preferably, in step 1.6, the spectral analysis is performed by performing a one-dimensional spectral analysis along each radial coordinate corresponding to each column of the transformed image block.

[0028] Preferably, in step 1.7, the fringe angle and / or fringe spacing and / or fringe density are determined based on the strongest frequency component of the power spectrum.

[0029] Preferably, the method further includes repeating steps 1.1 to 1.7 along the fracture surface and / or along the crack propagation path.

[0030] Preferably, the streak features along the fracture surface and / or crack propagation path are integrated to determine the integrated streak features and / or the error of the integrated streak features.

[0031] This invention provides a computer-implemented method for determining whether a part, particularly an aircraft part, has suffered fatigue damage, the method comprising:

[0032] - Perform the method according to any one of the preceding claims; and

[0033] - Using a fatigue damage determination device, the part is determined to have suffered fatigue damage when the peak value of the power spectrum exceeds a predetermined threshold, and the part is determined not to have suffered fatigue damage when the peak value does not exceed the predetermined threshold.

[0034] The present invention provides a system configured to perform a preferred method, the system comprising: a scanning microscope apparatus adapted to image the sample surface to obtain a sample image that may contain the fatigue striations; and a data processing device having:

[0035] - An image block selection device, which is adapted to select sample image blocks from the sample image that may contain the fatigue fringes;

[0036] - A preprocessing device adapted to normalize the sample image patch and enhance the linear regular structure contained in the sample image patch to obtain a normalized image patch.

[0037] - An autocorrelation device, which is adapted to determine the autocorrelation of the normalized image patch to obtain an autocorrelation image patch;

[0038] - A Radon transform apparatus, which is adapted to perform a Radon transform on the autocorrelation image block to obtain a transformed image block;

[0039] - A frequency analysis device adapted to perform spectral analysis on the transformed image block and determine the power spectrum of the transformed image block;

[0040] - A fringing feature determining device, the fringing feature determining device being adapted to determine at least one fringing feature suitable for indicating the presence of fatigue fringes in the sample image based on the power spectrum.

[0041] Preferably, the data processing device further includes a fatigue damage determination device adapted to determine at least one fringe feature in the fatigue fringes contained in the sample image.

[0042] The present invention provides a computer program product including instructions that, when executed by a computer, cause the computer to perform the method according to any one of the preceding claims.

[0043] This invention provides a computer-readable data carrier on which a preferred computer program product is stored. For example, the computer-readable data carrier may be a computer-readable data storage medium.

[0044] The present invention provides a data carrier signal, wherein the data carrier signal carries a preferred computer program product.

[0045] This invention is based on the technical fields of data science (computer vision, image processing, and pattern recognition), such as its application in materials science, particularly in materials failure analysis and fracture analysis. One idea is to determine fatigue striation characteristics, which are used in the failure analysis process to analyze the fracture surface of materials failing due to fatigue.

[0046] To determine whether a part has suffered fatigue damage and is prone to material failure, different fatigue striation characteristics will be identified, such as striation number, striation density, and striation angle. Another basic idea is to process a large number of fracture sites exhibiting fatigue striations and determine the striation characteristics of these fatigue striations, such as striation density (the number of striations per unit length). The striation characteristics along the crack propagation path of the fracture are integrated to estimate the total striation count.

[0047] The purpose of the methods and apparatus described herein is to allow robust estimation or determination of fringe density and orientation appearing in scanning electron microscope (SEM) images. Robust determination based on computer vision techniques is challenging due to the significant variations in the appearance of fringes in SEM images. Furthermore, the limited availability of training data in terms of both quality and quantity hinders the application of known supervised machine learning-based computer vision techniques.

[0048] The idea here is to provide an unsupervised, robust computer vision method for estimating or determining fringes density and fringes orientation, as a technical means for partially or fully automated fatigue fringes measurement systems.

[0049] First, at least one SEM image of the part to be tested is captured. The input to be processed is an image patch, i.e., a sub-part of the SEM image, representing the portion of the image to be analyzed. The goal is to robustly estimate the fringe density and fringe orientation within the image patch. This is particularly challenging when the image patch contains only blurred fringes.

[0050] As a first step, the input image patches are normalized and their linear, regular structure is enhanced. This can be achieved using standard image processing techniques, such as:

[0051] - Gaussian filtering operation, used to reduce image noise (see cf[1]).

[0052] - Histogram equalization (see cf[3])

[0053] - Image binarization (see cf[4])

[0054] - Morphological operations on binarized images, such as erosion, dilation, or skeletonization operations (see cf[1]).

[0055] - Self-filtering (see cf[5])

[0056] - Windowing of image patches to reduce the intensity at the edges of the patches (see cf[1])

[0057] - Normalize the intensity values ​​to a fixed range, for example, 0 to 1.

[0058] Following the preprocessing step, a normalized image representation with enhanced regular structure is obtained. The normalized image is further processed by calculating the autocorrelation of the preprocessed image patches, i.e., the correlation between the image patch and itself. Zero-padding image patches can be used as input in the frequency domain to efficiently compute the autocorrelation, thus avoiding the wrap-around effect and improving spatial resolution.

[0059] Then, a spectral frequency analysis of the previous autocorrelation is performed using the Radon transform.

[0060] A Radon transform is applied to the autocorrelation image to produce a transformed image, where the horizontal axis corresponds to the Radon projection angle and the vertical axis corresponds to the radial coordinate of the Radon projection.

[0061] Therefore, each column in the Radon transform corresponds to an intensity profile formed by integrating (projecting) the intensity of the autocorrelated image along the axis at a given angle.

[0062] Typically, only a small horizontal range (i.e., an angular range) is sufficient to represent the repeating structure of the original autocorrelation image. Since projection only along right angles does not erase the repeating structure of the autocorrelation image, the corresponding angle represents the fringe angle being searched. This makes the analysis highly selective for the correct fringe angle.

[0063] To identify columns with the strongest repeating structure, a standard one-dimensional spectral analysis, such as one based on Fourier Transform (FT) or Fast Fourier Transform (FFT), is applied to each column of the Radon transform. In this way, the column with the strongest frequency component in the Fourier spectrum (e.g., the highest value in the power spectrum) can be selected to determine the fringe angle being searched. By considering the known relationship between frequency and wavelength, knowing the strongest frequency component also yields the fringe spacing (the wavelength of the fringe). The reciprocal of the fringe spacing is the fringe density being searched.

[0064] If a frequency component corresponds to a meaningful number of fringe lines in the original input block (e.g., at least 5 fringe lines), the robustness of the process can be further improved by considering only those frequency components. In this way, if a strong frequency component in the Fourier spectrum corresponds to, for example, only 2 fringe lines in the input block, that strong frequency component will be ignored (e.g., by setting its amplitude to 0).

[0065] Applying a one-dimensional FFT to each column of the Radon transform allows for the computation of the power spectrum (the squared amplitude of the Fourier spectrum). As previously discussed, uncorrelated frequency components are set to 0. After the FFT, the horizontal axis remains angle (as in the Radon transform), while the vertical axis now represents spatial frequencies. The dominant frequencies of the fringe and their corresponding angles can usually be clearly identified by well-located peaks in the power spectrum (e.g., by finding the location of the maximum value of the power spectrum). This allows for robust identification of fringe spacing and fringe angles.

[0066] The maximum value of the determined power spectrum can further serve as an evidentiary measure for the presence of fringes in the input block. This evidentiary measure can be used to detect fringed regions within the full SEM image by performing the previously discussed steps (i.e., cutting out and analyzing many different blocks at different image locations) at different locations within the SEM image. If the corresponding block of a fringe provides an evidentiary measure higher than a given threshold, the fringe is considered to have been detected at the given image location.

[0067] The size of the input block (i.e., the width and height of the block, expressed in pixels) affects the quality of estimations of fringe density and fringe orientation. A block size that is too small may contain too few fringes, thus leading to inaccurate determinations of the spatial frequencies of the fringes. On the other hand, a block size that is too large may already include fracture regions that do not represent fringes, potentially introducing noise into the density and orientation estimates. The ideal block size can be determined by analyzing blocks of various sizes centered at a single point within the SEM image and subsequently selecting the block size with the highest measure of evidence (e.g., power spectral density).

[0068] Using the methods and apparatus described herein, fatigue fringes within SEM images can be identified or detected, and fringe density and orientation can be robustly determined, enabling highly automated fringe counting systems. Large numbers of SEM images with fracture surfaces (e.g., long crack propagation paths) can be automatically recorded. SEM images containing fatigue fringes can be automatically identified, and their center locations and ideal block sizes can be recorded.

[0069] In addition, the fringes density and fringes orientation at all recorded fringes locations can be automatically measured and recorded.

[0070] It can automatically filter recorded measurements that are inconsistent with known crack propagation physics based on estimated fringing orientation angles or fringing densities, thereby removing possible measurement outliers.

[0071] Furthermore, the fringing density along the crack propagation path can be automatically integrated to obtain and record the final fringing count. Statistical analysis of the large number of available measurements (e.g., the standard deviation of the fringing density for each measured SEM image) can be used to extend the measurement error to the error estimate for the final fringing count.

[0072] These results can be presented (e.g., in the form of a graphical user interface) to domain experts for verification or possible correction of automated processing steps. All corrections made by the domain expert can be recorded. Corrections may include removing erroneous automated measurements or adding new measurements at other relevant locations supported by the automated estimation of density and orientation (e.g., at click locations within the SEM image).

[0073] In addition, reports can be automatically generated based on the recorded measurement data.

[0074] Validation and calibration records performed by domain experts can be used to train machine learning algorithms, which may help avoid long-term manual calibration (e.g., by using validated automated measurements as positive training examples and calibrated automated measurements as negative training examples). Attached Figure Description

[0075] Embodiments of the invention will then be described in more detail with reference to the accompanying schematic diagrams. Wherein:

[0076] Figure 1 An example of a system for determining the characteristics of striated patterns is described;

[0077] Figure 2 An embodiment of a method for determining the presence of fatigue damage is described;

[0078] Figure 3 Examples depicting sample images and sample image patches;

[0079] Figure 4 An example depicting a normalized image patch;

[0080] Figure 5 An example depicting autocorrelation image patches;

[0081] Figure 6 An example of a transformed image patch is depicted; and

[0082] Figure 7 An example depicting the power spectrum is shown. Detailed Implementation

[0083] Figure 1 A system 10 for determining the presence of fatigue damage in part 12 is described.

[0084] System 10 includes an imaging device 14 and an evaluation device 16.

[0085] Imaging device 14 preferably includes scanning microscope device 18, such as scanning electron microscope (SEM).

[0086] The part 12 is positioned so that the imaging device 14 can capture an image of the sample surface 20 of the part 12.

[0087] The evaluation apparatus 16 preferably includes a data processing device 22, such as a computer. It should be noted that the data processing device 22 may be a single device or a combination of multiple devices to perform the steps of the determination method described below.

[0088] Figure 2 An example of a method for determining the presence of fatigue damage in part 12 is described.

[0089] like Figure 2 As depicted, firstly, imaging step S10 is performed using imaging device 14 (e.g., scanning microscope device 18). During imaging step S10, the sample surface 20 is scanned to obtain sample image 24. Figure 3 An example of sample image 24 generated due to the execution of imaging step S10 is depicted.

[0090] like Figure 3 As can be seen, sample image 24 contains multiple fatigue fringes 26.

[0091] The sample image 24 is acquired as a grayscale image from the scanning microscope apparatus 18 in a conventional manner. During subsequent processing, the sample image 24 can be considered as a numerical matrix arranged in rows and columns. In this matrix, a cell typically represents a single pixel of the sample image 24. Therefore, if a row or column of pixels is referenced, it can also be considered as a row or column representing the numbers of the corresponding pixels.

[0092] The sample image 24 is passed to the image block selection step S12. In this step, a sample image block 28 (e.g., a portion of the sample image 24) is selected using an image block selection device for further processing. The image block selection step S12 can be repeated several times on the same sample image 24 to select different portions as additional sample image blocks 30. Sample image block 28 (e.g., a portion of the sample image 24) Figure 3 The illustrated sample image blocks 30 and the other sample image blocks 30 are selected such that they are centered on the same pixel 29 of the sample image 24.

[0093] Sample image blocks 28 and 30 are passed to preprocessing step S14. In preprocessing step S14, sample image block 28 is normalized and the linear regular structure 32 is enhanced. The result of preprocessing step S14 is... Figure 4The normalized image blocks 34 depicted in the image are then depicted as negatives to make them easier to illustrate in grayscale.

[0094] The preprocessing step S14 is performed based on image processing techniques known in the art. In a simple example, a Gaussian filter may be applied to the sample image block 28 to reduce noise, and then the intensity values ​​of the sample image block 28 may be normalized to a fixed range from 0 to 1.

[0095] like Figure 4 As shown, the regular structure 32 exhibits a pattern of equidistant lines repeating at a certain dominant frequency. The window function is also visible and causes the regular structure 32 to fade as the radial spacing from the center toward the edges of the normalized image patch 34 increases.

[0096] The normalized image patch 34 is passed to the autocorrelation step S16. In the autocorrelation step S16, the autocorrelation of the normalized image patch 34 is calculated. In other words, the correlation between the normalized image patch and itself is calculated. The result of the autocorrelation step S16 is as follows: Figure 5 The autocorrelation image patch 36 depicted in the image. Due to autocorrelation, the periodicity of the features is further enhanced and indicated by the bar-shaped features 35.

[0097] The autocorrelated image block 36 is passed to the Radon transform step S18. In the Radon transform step S18, a Radon transform is performed on the autocorrelated image block 36 to produce a transformed image block 38. The Radon transform can be performed such that the horizontal axis corresponds to the Radon projection angle and the vertical axis corresponds to the radial component of the Radon projection. Figure 6 As depicted in this example, there are repeating structures 40 that form vertical lines close to the center of the transformed image patch 38. The paired vertical lines are added manually to aid in the identification of the repeating structures 40, and are not caused by any step of the method.

[0098] The transformed image block 38 is passed to the frequency analysis step S20. Here, a spectral analysis is performed on the transformed image block 38 to obtain the power spectrum 42. For example, a Fourier transform, preferably a Fast Fourier Transform or FFT, is performed on the transformed image block 38 and each column of the transformed image block 38.

[0099] Therefore, to obtain Figure 7 The power spectrum 42 is depicted in the figure. The power spectrum 42 is again the Radon projection angle on its horizontal axis. However, since there is only one periodicity in the current sample surface 20, the vertical axis is transformed from the spatial domain to the spatial frequency domain, thus producing a single peak 43 in this case.

[0100] The power spectrum 42 is transferred to the fringe feature determination step S22. In this step, the fringe features of the fatigue fringes 26 are obtained from the power spectrum 42. As previously mentioned, the position of the peak 43 of the power spectrum 42 includes the relevant fringe features. The horizontal position of the peak 43 from the origin corresponds to the angle at which the fatigue fringes 26 are aligned.

[0101] Furthermore, the position of the peak 43 of the power spectrum 42 along the vertical axis corresponds to the spatial frequency or "fringe per unit length" and is referred to as the fringe density. Additionally, in the fringe feature determination step S22, the results of another sample image block 30 can be processed to obtain an estimate of the fringe feature error determined based on the sample image block 28.

[0102] Optionally, a closing operation step S24 can be performed, which can be repeated along the crack path and along the crack 44 in part 12.

[0103] It should be noted that steps S10 to S22, and optionally step S24, form a method for determining the characteristics of flare.

[0104] The method for determining the presence of fatigue damage further includes a fatigue damage determination step S26. In this step, the results of the measurement methods S10 to S24 are processed. Basically, in the fatigue damage determination step S26, if the peak value of the power spectrum exceeds a predetermined threshold, it is determined that part 12 has suffered fatigue damage. If the peak value does not exceed the predetermined threshold, it is determined that part 12 has not suffered fatigue damage. Furthermore, in step S26, measurement anomalies or measurement results that are inconsistent with known crack propagation physics are discarded. In addition, if "gaps" are left due to the discarded measurement data, step S26 may prompt the measurement method to be repeated to fill these "gaps".

[0105] To improve the determination or assessment of fatigue damage to parts, a computer-implemented method for determining the fringing characteristics of fatigue fringes (26) on a sample surface of a part (12) is proposed. The sample surface (20) is imaged using a scanning electron microscope to obtain a sample image (24) that may contain the fatigue fringes (26). A sample image patch (28) that may contain the fatigue fringes (26) is selected from the sample image (24) for further processing. After normalizing the sample image patch (28) and enhancing the linear regularity contained within it, the resulting normalized image patch (34) is subjected to autocorrelation, Lardon transform, and spectral analysis. If any fatigue fringes (26) are present, the resulting power spectrum (42) of the transformed image patch (38) contains information about the fringing characteristics of the fatigue fringes (26) contained in the sample image (24). Furthermore, a system for performing the method is proposed.

[0106] List of reference numerals

[0107] 10 System

[0108] 12 parts

[0109] 14 Imaging devices

[0110] 16 Evaluation Device

[0111] 18. Scanning microscope setup

[0112] 20 Sample surfaces

[0113] 22 Data processing device

[0114] 24 sample images

[0115] 26 Fatigue streaks

[0116] 28 sample image patches

[0117] 29 pixels

[0118] 30 Another sample image patch

[0119] 32. Rule Structure

[0120] 34 Normalized image patches

[0121] 35 Strip-shaped features

[0122] 36 Autocorrelation Image Patch

[0123] 38 Transformed image blocks

[0124] 40. Repeated Structures

[0125] 42 Power Spectrum

[0126] 43 peak value

[0127] 44 Cracks

[0128] S10 Imaging Steps

[0129] S12 Image Patch Selection Steps

[0130] S14 Preprocessing Steps

[0131] S16 Autocorrelation Steps

[0132] S18 Ladon Transformation Steps

[0133] S20 Frequency Analysis Steps

[0134] S22 Steps for determining the characteristics of striated patterns

[0135] S24 Closing Operation Steps

[0136] S26 Fatigue Damage Determination Steps

Claims

1. A computer-implemented method for determining fatigue striations (26) on a sample surface of a part (12), the method comprising: Step 1.1 Using a scanning microscope (18), image the sample surface (20) to obtain a sample image (24) that may contain the fatigue striations (26); Step 1.2 Using the image block selection device, select sample image blocks (28) from the sample image (24) that may contain the fatigue fringe (26) for further processing; Step 1.3 Using a preprocessing device, the sample image block (28) is normalized and the linear regular structure contained in the sample image block (28) is enhanced to obtain a normalized image block (34); Step 1.4 Use an autocorrelation device to determine the autocorrelation of the normalized image block (34) to obtain an autocorrelation image block (36); Step 1.5 Using the Radon transform device, perform Radon transform on the autocorrelation image block (36) to obtain the transformed image block (38); Step 1.6 Using a frequency analysis device, perform spectrum analysis on the transformed image block (38) and determine the power spectrum (42) of the transformed image block (38); Step 1.7 Using the fringe feature determination device, determine at least one fringe feature suitable for indicating the presence of fatigue fringe (26) in the sample image block (28) based on the power spectrum (42).

2. The method according to claim 1, wherein, In step 1.1, the scanning microscope device (18) is selected from the group consisting of scanning electron microscopes or scanning probe microscopes.

3. The method according to claim 1, wherein, In step 1.2, multiple sample image blocks (28) of different sizes are selected, and the sample image blocks (28) are centered on the same point of the sample image (24), and steps 1.3 to 1.7 are performed on each of the sample image blocks (28).

4. The method according to claim 1, wherein, Step 1.3 includes performing any of the following operations on the sample image block (28): Step 4.1 Gaussian filtering operation; and / or Step 4.2 Histogram equalization; and / or Step 4.3 Image binarization and subsequent morphological operations, such as erosion, dilation, or skeletonization operations; and / or Step 4.4 Frequency domain self-filtering operation; and / or Step 4.5 involves window operations, such as applying the Hamming window; and The following operations are performed on the sample image block (28) to obtain the normalized image block (34): Step 4.6 Normalization operation is used to normalize the intensity values ​​to a fixed range.

5. The method according to claim 1, wherein, In step 1.4, the autocorrelation is determined via the frequency domain using zero-padding input image blocks.

6. The method according to claim 1, wherein, In step 1.6, the spectral analysis is performed by performing a one-dimensional spectral analysis along each radial coordinate corresponding to each column of the transformed image block (38).

7. The method according to claim 1, wherein, In step 1.7, the fringe angle and / or fringe spacing and / or fringe density are determined based on the strongest frequency component of the power spectrum (42).

8. The method of claim 1, further comprising repeating steps 1.1 to 1.7 along the fracture surface and / or along the crack propagation path.

9. The method according to claim 8, wherein, Integrate the streak features along the fracture surface and / or crack propagation path, and thereby determine the integrated streak features and / or the error of the integrated streak features.

10. A computer-implemented method for determining whether a part has suffered fatigue damage, the method comprising: Step 10.1 Perform the method according to any one of the preceding claims; as well as Step 10.2 Using a fatigue damage determination device, the part is determined to have suffered fatigue damage when the peak value (43) of the power spectrum (42) exceeds a predetermined threshold, and the part is determined not to have suffered fatigue damage when the peak value (43) does not exceed the predetermined threshold.

11. A system configured to perform the method according to any one of claims 1-10, the system comprising: A scanning microscope apparatus adapted to image the sample surface (20) to obtain a sample image (24) that may contain the fatigue striations (26); and a data processing device having: - Image block selection device, which is adapted to select sample image (24) blocks that may contain the fatigue fringe (26) from the sample image (24); - A preprocessing device adapted to normalize the sample image (24) block and enhance the linear regular structure (32) contained in the sample image block (28) to obtain a normalized image block; - An autocorrelation device adapted to determine the autocorrelation of the normalized image patch to obtain an autocorrelation image patch (36); - A Radon transform apparatus adapted to perform a Radon transform on the autocorrelation image block (36) to obtain a transformed image block (38); - A frequency analysis device adapted to perform spectral analysis on the transformed image block (38) and determine the power spectrum (42) of the transformed image block (38); - A fringing feature determining device, the fringing feature determining device being adapted to determine, based on the power spectrum (42), at least one fringing feature suitable for indicating the presence of fatigue fringes (26) in the sample image block (28); and - The data processing device has a fatigue damage determination device, which is adapted to determine that the part has suffered fatigue damage when the peak value (43) of the power spectrum (42) exceeds a predetermined threshold, and to determine that the part has not suffered fatigue damage when the peak value (43) does not exceed the predetermined threshold.

12. A computer-readable data storage medium having a computer program containing instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 10.

Images